An ar technology-based construction auxiliary guidance method and system

Abstract

The application discloses a kind of AR technology-based construction auxiliary guidance method and system, it is related to building information technology technical field, wherein including: obtaining building geometry and semantic information by BIM model, establish global coordinate and calculate anchor point, construct BIM-semantic topological pose graph;Obtain structural plane feature, anchor point matching and pose solution are carried out in combination with BIM-semantic topological pose graph, obtain global initial pose;Real-time global accurate pose is solved using hierarchical pose graph optimization algorithm;According to global accurate pose, concealed engineering component model in BIM construction model is superimposed and rendered to AR terminal picture;When floor switching event is identified, trigger building level global pose graph optimization, and maintain the coordinate system continuity of AR superimposed content, carry out no-sense switching.The application aims to solve the technical problem that large-scale global space consistency is difficult to guarantee under complex building environment.

Description

Technical Field

[0001] This invention relates to the field of building information technology, and in particular to a building construction assistance guidance method and system based on AR technology. Background Technology

[0002] Augmented reality (AR) technology can significantly improve construction efficiency and quality by overlaying digital information such as Building Information Modeling (BIM) onto the real construction site. However, in large-scale and complex scenarios such as super high-rise buildings and large industrial parks, existing AR construction assistance methods face the technical bottleneck of difficulty in ensuring global spatial consistency. Specifically: on the one hand, during long-term, large-scale mapping, traditional SLAM methods suffer from a lack of effective global loop closure constraints, resulting in accumulated errors that continuously increase with the exploration path. Furthermore, due to the highly similar structural features between different floors, loop closure detection errors are easily triggered, leading to severe map distortion or even system crashes. On the other hand, super high-rise buildings contain dozens or even hundreds of floors, and large industrial parks encompass multiple independent work areas. Each area is usually mapped independently, and the transformation relationship between their local coordinate systems is difficult to accurately define, resulting in AR content across floors and work areas not being seamlessly connected in the same global coordinate system. While existing marker-based positioning methods can guarantee local accuracy, deploying markers layer by layer and area by area is time-consuming and costly to maintain, and cannot fundamentally solve the problem of globally consistent and unified representation in large-scale scenarios. Therefore, there is an urgent need for an AR-based construction assistance and guidance method that can balance large-scale global consistency with local positioning accuracy. Summary of the Invention

[0003] This invention provides a construction assistance guidance method based on AR technology, comprising: By acquiring architectural geometric and semantic information from the BIM model, establishing global coordinates and calculating anchor points, a BIM-semantic topological pose diagram is constructed. The structural planar features are obtained by synchronous positioning and mapping system, and anchor point matching and pose calculation are performed by combining BIM-semantic topology pose graph to obtain global initial pose. Based on the BIM-semantic topological pose graph and the global initial pose, the real-time global accurate pose is calculated using a hierarchical pose graph optimization algorithm. Based on the global precise pose, the hidden engineering component models in the BIM construction model are superimposed and rendered onto the AR terminal screen; When a floor switching event is detected, the building-level global pose graph is optimized, and the coordinate system continuity of the AR overlay content is maintained to achieve a seamless switching.

[0004] The aforementioned AR-based construction assistance guidance method acquires building geometric and semantic information from a BIM model, establishes global coordinates and calculates anchor points, and constructs a BIM-semantic topological pose graph, including: Analyze the BIM model and extract the geometric structure information of each floor; Based on the extracted geometric information, the spatial hierarchical semantics and topological connections of the organizational structure tree are extracted; Establish a global coordinate system and calculate the global coordinates of BIM anchor points for each floor. Calculate the anchor points corresponding to the vertical passages running through different floors, and construct a BIM-semantic topology pose graph.

[0005] The aforementioned AR-based construction assistance guidance method acquires structural planar features through a synchronous positioning and mapping system, and combines BIM-semantic topological pose graphs for anchor point matching and pose calculation to obtain a global initial pose, including: The multi-sensor fusion SLAM system of the AR terminal device is activated to collect environmental data in real time and build a local point cloud map, while outputting the initial tracking pose of the device in the SLAM local coordinate system. Structural planar features are extracted from the real-time constructed local point cloud map, and the intersection lines and intersection points between the planes are calculated; The extracted structural planar features are matched online with the anchor point geometric descriptions of the current floor in the BIM-semantic topological pose map. By utilizing the successfully matched anchor point correspondence, the global initial pose of the device in the global coordinate system is obtained by solving the six-degree-of-freedom pose of the device.

[0006] The aforementioned AR-based construction assistance guidance method activates the multi-sensor fusion SLAM system of the AR terminal device, collects environmental data in real time and constructs a local point cloud map, and simultaneously outputs the initial tracking pose of the device in the SLAM local coordinate system, including: Initialize the multi-sensor data stream and complete the time synchronization and spatial extrinsic parameter calibration of the lidar, inertial measurement unit and camera; Run the laser-inertial-vision tightly coupled SLAM front-end odometry calculation method to output the device's initial tracking pose and local dense point cloud map in the SLAM local coordinate system.

[0007] The aforementioned AR-based construction assistance guidance method, based on a BIM-semantic topological pose graph and a global initial pose, utilizes a hierarchical pose graph optimization algorithm to calculate the real-time global accurate pose, including: The global initial pose is used as the reference node of the pose graph, and new pose nodes, odometry constraint edges and BIM anchor point constraint edges are dynamically added during the continuous SLAM tracking process to construct a real-time full pose graph that integrates BIM anchor point constraints. An anchor-constraint-driven hierarchical pose graph optimization algorithm is used to perform progressive global optimization of the full pose graph at the local window level, floor level, and building level to obtain the global accurate pose.

[0008] The aforementioned AR-based construction assistance guidance method uses the global initial pose as the reference node of the pose graph, and dynamically adds new pose nodes, odometry constraint edges, and BIM anchor point constraint edges during continuous SLAM tracking to construct a real-time full pose graph incorporating BIM anchor point constraints, including: The global initial pose is used as the first reference node of the pose graph to initialize the full pose graph structure; As the SLAM odometry continuously outputs new poses, new pose nodes are dynamically created, and odometry constraint edges are added between adjacent nodes. During SLAM tracking, structural plane feature extraction and BIM anchor point matching are continuously performed. After a successful match, BIM anchor point constraint edges are generated, and the information matrix of the constraint edge is dynamically calculated based on the matching score and injected into the pose graph.

[0009] The aforementioned AR-based construction assistance guidance method utilizes an anchor-constraint-driven hierarchical pose graph optimization algorithm to perform progressive global optimization of the entire pose graph at the local window level, floor level, and building level, resulting in a globally accurate pose. This includes: Perform high-frequency local optimization on the pose nodes within the current sliding window to instantly eliminate short-term tracking jitter; When the triggering condition is met, all pose nodes within the current floor are uniformly optimized to eliminate the cumulative drift error of long-distance movement; When a floor switching event is detected, the cross-floor constraint of the through anchor point is used to perform joint global optimization on all traversed floors and unify the cross-floor coordinate system.

[0010] An AR-based construction assistance and guidance system includes: The BIM pose graph construction module is used to obtain building geometry and semantic information from the BIM model, establish global coordinates and calculate anchor points to construct a BIM-semantic topology pose graph. The global initial localization module is used to obtain structural planar features through the synchronous localization and mapping system, and combine them with the BIM-semantic topological pose graph to perform anchor point matching and pose calculation to obtain the global initial pose. The hierarchical pose optimization module is used to calculate the real-time global accurate pose based on the BIM-semantic topology pose graph and the global initial pose using the hierarchical pose graph optimization algorithm. The AR overlay rendering module is used to overlay and render the hidden engineering component models in the BIM construction model onto the AR terminal screen based on the global precise pose. The seamless cross-floor switching module is used to trigger building-level global pose graph optimization when a floor switching event is detected, and to maintain the coordinate system continuity of the AR overlay content for seamless switching.

[0011] The beneficial effects achieved by this invention are as follows: It achieves global spatial consistency positioning in large-scale complex scenes such as super high-rise buildings and large industrial parks, and completely eliminates the coordinate system fragmentation and drift problem of AR content across floors and work areas; By adopting a hierarchical pose graph optimization strategy, the computational complexity of global optimization is reduced from O(N²) to O(F×K²), which significantly improves the real-time performance and robustness in large-scale scenarios. It eliminates the need to deploy a large number of physical markers layer by layer and area by area, reducing on-site deployment costs and achieving a balance between high-precision positioning and lightweight deployment. It can be widely used in positioning and setting out, quality verification and progress management in building construction. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1 This is a flowchart of a construction assistance guidance method based on AR technology provided in Embodiment 1 of this application.

[0014] Figure 2 This is a schematic diagram of a construction assistance guidance system based on AR technology provided in Embodiment 2 of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1

[0017] like Figure 1As shown, Embodiment 1 of this application provides a construction assistance guidance method based on AR technology, including: S1: Obtain building geometry and semantic information through BIM model, establish global coordinates and calculate anchor points, and construct BIM-semantic topology pose graph; The geometric structure information and semantic topological relationship of the building to be constructed are obtained through the BIM model. A global coordinate system with the preset benchmark point on the first floor of the building as the origin is established and the BIM anchor point coordinates of each floor are calculated. A BIM-semantic topological pose diagram containing a global coordinate reference system and a multi-level anchor point hierarchy is constructed.

[0018] The process of acquiring architectural geometric and semantic information from the BIM model, establishing global coordinates and calculating anchor points, and constructing a BIM-semantic topological pose graph includes the following sub-steps: S11: Analyze the BIM model and extract the geometric structure information of each floor; Obtain the BIM model file of the building project to be constructed, in IFC or Revit native format. Use the BIM parsing engine to read the complete data hierarchy of the model, identifying the organizational structure tree of "Project—Site—Building—Floor—Component," thus establishing an index foundation for subsequent layer-by-layer traversal and information extraction.

[0019] For components categorized as "wall," "column," and "floor slab" in the BIM model, their spatial location and geometric parameters are extracted. For wall components, their start-point coordinates, end-point coordinates, wall thickness, and height are extracted, and the equation of the wall's central axis is calculated. For column members, the coordinates of their cross-section center point, cross-section dimensions, and column height are extracted; for floor slab members, their bottom elevation, top elevation, and planar outline boundary are extracted. All of the above geometric data are recorded using the BIM model's internal coordinate system and stored as structured data tables.

[0020] Identify components in the BIM model that are of IFC type "staircase" and "elevator", and extract their planar outline boundaries and vertical elevation ranges. For stairwells, record the coordinates of the entrance plane corner points and the turning points of the stair platforms on each floor; for elevator shafts, record the starting and ending coordinates of the complete vertical axis from the first floor to the top floor, and mark the plane intersection points of this axis on each floor.

[0021] S12: Based on the extracted geometric information, extract the spatial hierarchical semantics and topological connections of the organizational structure tree; By leveraging the association attributes between components and spaces in the BIM model, each wall component is assigned to its corresponding room space, and each room space is then assigned to its corresponding floor. If the BIM model lacks explicit room space definitions, a planar segmentation algorithm is used to identify closed polygonal areas enclosed by walls within the same floor as room spaces, and the attribution relationship between walls and rooms is automatically established. Ultimately, a semantic hierarchy tree is constructed with "Building" as the root node, "Floor" and "Room" as intermediate nodes, and "Wall / Column" as leaf nodes.

[0022] For each floor, the intersection of the column grid and the L-shaped / T-shaped intersection of the wall axis are used as candidate anchor points. Based on the actual wall connection paths between each candidate point, topological adjacency relationships such as "adjacent", "shared wall", and "diagonal" are established and stored in the form of an undirected graph. The nodes of the graph are candidate anchor points, and the edges of the graph represent direct spatial connection paths between two points.

[0023] S13: Establish the global coordinate system and calculate the global coordinates of BIM anchor points for each floor; The corner point of the stairwell on the first floor of the building is selected as the origin of the global coordinate system. A right-handed rectangular global coordinate system is established with the east direction of the first floor as the positive X-axis, the north direction as the positive Y-axis, and the vertical upward direction as the positive Z-axis. If the coordinate system inside the BIM model is inconsistent with the selected global coordinate system, the rotation and translation transformation matrix from the BIM model coordinate system to the global coordinate system is calculated and stored.

[0024] For each floor, the intersection of components that meets one of the following conditions will be selected as the BIM anchor point for that floor: The anchor points are selected based on the following criteria: intersection of column grid axes, L-shaped / T-shaped intersections of load-bearing walls, and corner points of stairwell entrances. The priority for anchor point selection is: column grid intersections are preferred over stairwell corner points, and stairwell corner points are preferred over load-bearing wall intersections.

[0025] After selection, the internal coordinates of each anchor point in the BIM model are converted into three-dimensional coordinates in the global coordinate system using rotation and translation transformation matrices, denoted as... ,in That is, the floor elevation where the anchor point is located.

[0026] S14: Calculate the anchor points corresponding to the vertical passages running through different floors and construct the BIM-semantic topology pose graph.

[0027] Traverse the data of stairwells and elevator shafts, and mark the plane corners of each vertical through structure on each floor as through anchor points.

[0028] For elevator shafts, the intersections of their through-axis with the planes of each floor are also marked as through-anchor points. For the same vertical through-structure, a "cross-floor connection edge" is established between the corresponding through-anchor points on adjacent floors. This edge records the spatial transformation relationship between the upper and lower anchor points in the global coordinate system: since the through-structure is strictly continuous in the vertical direction, the translation component of this transformation relationship only has the floor height difference in the Z direction, while the rotation component is zero. This provides cross-floor rigid constraints for the building-level global optimization in subsequent steps.

[0029] A three-layer graph structure is then constructed. The top layer is the building layer, which contains overall building information; the middle layer consists of floor nodes, each containing the floor's elevation, plan range, and pointers to all BIM anchor points for that floor; the bottom layer consists of anchor node nodes, each storing its 3D coordinates in the global coordinate system, its room label, a list of topological adjacency edges with other anchor points within that floor, and pointers to connection edges with corresponding anchor points across floors.

[0030] The graph structure is serialized into a binary or XML file for subsequent loading and use on AR terminal devices.

[0031] S2: Obtain structural planar features through a synchronous positioning and mapping system, and combine them with a BIM-semantic topological pose graph to perform anchor point matching and pose calculation to obtain the global initial pose; Real-time environmental perception and local point cloud map construction are performed using a SLAM system. Structural planar features are extracted from the real-time point cloud and matched online with the anchor point geometric description of the current floor in the BIM-semantic topological pose map. The initial six-degree-of-freedom pose in the global coordinate system is obtained by solving the spatial transformation relationship of the matched anchor point pairs.

[0032] The process involves obtaining structural planar features through a synchronous localization and mapping (SMR) system, and then performing anchor point matching and pose calculation using a BIM-semantic topological pose graph to obtain the global initial pose. This includes the following sub-steps: S21: Start the multi-sensor fusion SLAM system of the AR terminal device, collect environmental data in real time and build a local point cloud map, and output the initial tracking pose of the device in the SLAM local coordinate system. The process of activating the multi-sensor fusion SLAM system of the AR terminal device, collecting environmental data in real time and constructing a local dense point cloud map, and outputting the initial tracking pose of the device in the SLAM local coordinate system includes the following sub-steps: S211: Initialize the multi-sensor data stream and complete the time synchronization and spatial extrinsic parameter calibration of the lidar, inertial measurement unit and camera; The SLAM system on the AR terminal device (such as AR smart glasses or handheld tablet) is activated on the target construction floor. First, the sensor extrinsic parameter files preset by the device at the factory or calibrated on site are loaded, and the spatial transformation matrix between the LiDAR and the camera is calibrated, as well as the timestamp alignment between the LiDAR and the inertial measurement unit.

[0033] After synchronization is complete, the system synchronously collects the following three types of data at a frequency of 10Hz to 30Hz: three-dimensional point cloud frames output by the lidar, three-axis acceleration and three-axis angular velocity output by the inertial measurement unit (IMU), and RGB image sequences output by the camera.

[0034] S212: Runs the laser-inertial-vision tightly coupled SLAM front-end odometry calculation method, outputting the device's initial tracking pose and local dense point cloud map in the SLAM local coordinate system.

[0035] First, the timestamps of the three sensor data collected at the current moment by the lidar, inertial measurement unit and camera are aligned. The lidar point cloud frames, camera image frames and IMU measurement values ​​with timestamp deviations less than a preset threshold are grouped into a set of synchronous observation data.

[0036] Next, the IMU measurements are pre-integrated between the previous frame and the current frame to calculate the relative position increment, velocity increment, and attitude change between the two frames, which serve as the inter-frame motion prior constraints for subsequent joint optimization.

[0037] Simultaneously, motion distortion correction is performed on the lidar point cloud frame. Utilizing the high-frequency motion estimation from the IMU pre-integration output, interpolation compensation is applied to the acquisition timestamp of each laser point within the current lidar point cloud frame, uniformly transforming all laser points to the coordinate system at the start of the frame, thus eliminating internal point cloud deformation caused by device movement.

[0038] After correction, feature extraction is performed on the laser point cloud. The local curvature of each laser point is calculated, and points in the point cloud are classified into corner points and planar points according to a preset curvature threshold: points with curvature greater than the high threshold are marked as corner points, corresponding to significant geometric features such as wall corners and column corners; points with curvature less than the low threshold are marked as planar points, corresponding to large-area planar structures such as walls and floors. Based on the characteristic of flat and continuous walls in the construction environment, the high threshold is set to an empirical value that can distinguish between edges and planes, and the low threshold is set to an empirical value that can retain complete planar information.

[0039] Subsequently, ORB feature point extraction is performed on the camera images. The number of feature points extracted is dynamically adjusted according to the image resolution to maintain 200 to 500 feature points per frame, so as to balance computational efficiency and robustness of feature matching.

[0040] After feature extraction is completed, the system constructs three types of residual terms. Specifically, the corrected laser corner points and planar points in the current frame are matched with the local map maintained by the system: for corner points, the nearest corner point is searched in the local map to construct the point-to-point distance residual; for planar points, the nearest plane is searched in the local map to construct the point-to-plane distance residual.

[0041] The ORB feature points extracted from the current frame image are matched with 3D map points in the local map, and the matching relationship is used to construct the visual reprojection residual. In a building construction environment, if the system detects that the number of feature points in the image is lower than a preset lower threshold, it indicates that the current area is a weak texture area such as a white wall. In this case, the system automatically lowers the weight coefficient of the visual residual of that frame to avoid low-quality observation contamination of the state estimation.

[0042] The IMU pre-integration residual is constructed based on the deviation between the inter-frame relative motion calculated by IMU pre-integration and the current state estimate.

[0043] Then, the laser residual, visual residual, and IMU pre-integration residual terms are constructed into a unified objective function. Within a sliding window of 10 to 20 keyframes, this objective function is optimized using nonlinear least squares, iteratively solving for the optimal pose and map point positions for all keyframes within the window.

[0044] During the optimization process, when the IMU detects that the device is in an emergency stop or emergency start state, the weight of the IMU pre-integrated residual is automatically increased to give priority to using inertial information to constrain inter-frame motion and make up for the response lag caused by the observation delay of laser and vision.

[0045] After optimization and convergence, the output device's initial tracking pose at the current moment is recorded in the SLAM local coordinate system with the device's location at the SLAM startup time as the origin and the device's orientation at the startup time as the initial direction; as well as the local dense point cloud map of the current frame after motion compensation.

[0046] Finally, it is determined whether the pose change or time interval between the current frame and the previous keyframe exceeds a preset threshold. If it does, the current frame is marked as a new keyframe, its pose node is added to the SLAM real-time pose graph, and its point cloud data is fused into the local map for matching and optimization in subsequent frames.

[0047] If the pose change or time interval between the current frame and the previous keyframe does not exceed a preset threshold, the system marks the current frame as a normal frame and does not add it to the SLAM real-time pose graph. However, its point cloud data is still used to update temporary observation information in the local map. Specifically, after registering the point cloud of the current frame with the local map, the system only retains the observation updates for existing planes in the local map and does not add new map points. At the same time, the system continues to use the pose node of the previous keyframe as the end node of the current sliding window, attaching the observation residual constraints of the current frame to this node to participate in the next round of sliding window optimization. Once the accumulated pose change or time interval meets the keyframe determination conditions, the system then formally inserts the latest keyframe into the SLAM real-time pose graph, completing the growth of pose graph nodes and the persistent update of the local map.

[0048] S22: Extract structural planar features from the real-time constructed local point cloud map and calculate the intersection lines and intersection points between the planes; For the local dense point cloud in the current frame, voxel filtering is first used for downsampling, with the voxel grid side length set to 0.05 meters to 0.1 meters. Then, a statistical filtering algorithm is used to remove outlier noise points. For the preprocessed point cloud, based on the similarity of the point's normal vectors and spatial proximity, a region growing algorithm is run to cluster points belonging to the same plane into a set of points within that plane. For each clustered set of points within the plane, a random sampling consensus algorithm or least squares method is used to fit the plane, obtaining the plane's normal vector and distance parameters to the origin. Based on the direction of the normal vector, wall planes (normal vectors approximately horizontal), ground planes (normal vectors approximately vertically upward), and cylindrical planes (normal vectors horizontal, and the point cloud distribution is arc-shaped, requiring further fitting of cylindrical parameters) are distinguished.

[0049] For any two extracted non-parallel planes, solve for their intersection equation by simultaneously solving the equations of the two planes. For three pairwise intersecting planes, solve for the coordinates of their common intersection point. Filter the intersection point type: the intersection of two perpendicular walls is a corner line, the intersection of a corner line and the ground is a corner point; the point where three surfaces intersect is a column corner point or a T-shaped / L-shaped wall intersection point. Record the filtered set of intersection points as the local structural feature point set of the current frame. Record the three-dimensional coordinates of each intersection point in the SLAM local coordinate system and attach its geometric type label (corner point, column corner point, wall intersection point).

[0050] S23: Perform online association matching between the extracted structural planar features and the anchor point geometric description of the current floor in the BIM-semantic topological pose map; Read the BIM-semantic topology pose map file. Based on the trend of the Z-axis coordinate change in the initial tracking pose, determine whether the device is in an ascending or descending state, and combine this with the design elevation information of each floor of the building to determine the floor number where the device is currently located.

[0051] Load the set of anchor points corresponding to the floor from the BIM-semantic topology pose graph. Each anchor point contains its three-dimensional coordinates in the global coordinate system, anchor point type label (column grid intersection, stairwell corner point, load-bearing wall intersection), and topological adjacency relationship between anchor points.

[0052] Subsequently, a matching score function is constructed to evaluate the pairwise matching assumptions between the set of local structural feature points and the set of BIM anchor points. The optimal matching pair is selected based on the comprehensive score of geometric consistency and topological consistency. The specific calculation formula is as follows: ,in, These are the weights for the geometric consistency score and the topological consistency score, respectively, for example... The geometric consistency score measures the alignment between the orthogonal directions at feature points and the orthogonal directions at design anchor points. It is calculated as the average of the absolute inner products, with a range of values ​​ranging from [value missing]. A larger value indicates a more consistent geometric orientation; ,in, and It is composed of feature points The unit vectors of the two lines of intersection determined by the two intersecting wall planes (for the corner point, these are the directions of the two wall lines and the corner line; for the corner point of the column, they are the directions of the lines of intersection of the tangent planes of the column) are approximately orthogonal. and BIM anchor point The unit vector of the design intersection direction of the two corresponding wall surfaces (or structural surfaces) is obtained from the wall axis direction of the BIM model; This is the topology consistency score item. This score measures the consistency between the local topology connection pattern centered on this point and the BIM design topology, and its value range is [value range missing]. A larger value indicates a better match in topology; ,in, For type matching indicator functions, when feature points Geometric type labels (such as "corner point", "column corner point", "wall T-intersection") and BIM anchor points The value is 1 when the type tags are completely identical, otherwise it is 0; Indicates in the local structure diagram and There exists a set of neighborhood feature points that are directly topologically connected (i.e. directly connected through the same wall or intersection line); This represents the number of neighboring points; Adjacency consistency indicator function, for feature points a neighboring point If in the BIM-semantic topology pose graph, the corresponding candidate anchor point and ( To and If there is a "neighboring" or "shared wall" topological connection between the corresponding candidate anchor point indices, the function takes the value 1; otherwise, it takes the value 0.

[0053] S24: Using the successfully matched anchor point correspondence, calculate the six-degree-of-freedom pose of the device in the global coordinate system to obtain the global initial pose.

[0054] The global initial pose problem is modeled as a least-squares optimization problem, with the goal of finding the optimal rotation matrix and translation vector such that the sum of the squared Euclidean distances between all matched feature points and their corresponding BIM anchor points after rotation and translation transformations is minimized. An iterative nearest-point solution method based on singular value decomposition is used to calculate the optimal transformation: first, the centroids of each matched point pair are calculated and removed, and a covariance matrix is ​​constructed. Singular value decomposition is then performed on the covariance matrix, and the optimal rotation matrix is ​​solved using the resulting orthogonal matrix. Finally, the optimal translation vector is calculated using the difference between the centroids of the two sets of points and the optimal rotation matrix.

[0055] Then, the root mean square error between all matched point pairs after transformation is calculated. If the root mean square error is less than a preset threshold (e.g., 0.05 meters) and the proportion of matched inliers (inliers are defined as point pairs with a distance deviation of less than 0.1 meters after transformation) is greater than a preset proportion (e.g., 70%), the pose calculation result is considered valid.

[0056] The effective pose calculation results are output in the form of a transformation matrix, which is the initial six-DOF pose of the device in the BIM global coordinate system. At the same time, based on this pose, all new pose nodes generated by subsequent SLAM tracking are transformed to the global coordinate system, completing the coordinate system migration of the SLAM real-time pose map from the local coordinate system to the global coordinate system, and obtaining the global initial pose.

[0057] S3: Based on the BIM-semantic topology pose graph and the global initial pose, the real-time global accurate pose is calculated using a hierarchical pose graph optimization algorithm; Based on the BIM-semantic topology pose graph and the global initial pose, a pose graph that integrates BIM anchor point constraints is continuously constructed during the SLAM real-time tracking process. Then, using the anchor point constraint-driven hierarchical pose graph optimization algorithm, a three-level progressive global optimization is performed sequentially at the local window level, floor level, and building level to obtain a real-time global accurate pose that is consistent across floors.

[0058] The process involves calculating the real-time global accurate pose using a hierarchical pose graph optimization algorithm based on the BIM-semantic topological pose graph and the global initial pose. This includes the following sub-steps: S31: Use the global initial pose as the reference node of the pose graph, and dynamically add new pose nodes, odometry constraint edges and BIM anchor point constraint edges during the continuous SLAM tracking process to construct a real-time full pose graph that integrates BIM anchor point constraints. The process involves using the global initial pose as the reference node for the pose graph, and dynamically adding new pose nodes, odometry constraint edges, and BIM anchor point constraint edges during continuous SLAM tracking to construct a real-time full pose graph incorporating BIM anchor point constraints. This includes the following sub-steps: S311: Initialize the full pose graph structure using the global initial pose as the first reference node of the pose graph; The global initial pose transformation matrix is ​​used as the first node of the pose graph. This node stores the six-DOF pose of the device in the BIM global coordinate system at startup. The pose is represented by a seven-dimensional vector, where the translation part is a three-dimensional vector and the rotation part is a unit quaternion to avoid the gimbaling problem of Euler angles.

[0059] The initial full pose graph consists of a node set, an odometry edge set, and an anchor constraint edge set. The node set initially contains only the reference node, and node attributes include a unique identifier, the floor number it belongs to, a seven-dimensional pose vector, and a timestamp. The odometry edge set and the anchor constraint edge set are initially empty.

[0060] The BIM anchor point set is managed according to an on-demand loading strategy. Specifically, during initialization, only the BIM anchor point nodes of the first floor of the building are loaded, and the fixed three-dimensional coordinates and anchor point type label of each BIM anchor point in the global coordinate system are stored as static reference nodes that do not participate in optimization iterations.

[0061] S312: When the SLAM odometry continuously outputs new poses, dynamically create new pose nodes and add odometry constraint edges between adjacent nodes.

[0062] When the SLAM system outputs a new device pose estimate at a frequency of 10Hz to 30Hz, keyframes are selected for each new pose frame. Specifically, the time interval between the current keyframe and the previous keyframe exceeds 0.5 seconds, or the spatial distance between the current keyframe and the previous keyframe exceeds 0.3 meters, or the rotation angle changes by more than 15 degrees.

[0063] Pose positions that meet the criteria are created as new pose nodes and added to the node set of the pose graph. The floor number to which the device belongs is determined based on the device's current height. When the device's Z-coordinate approaches the elevation of a certain floor, the BIM anchor points of that floor and adjacent floors are dynamically loaded as needed, while maintaining anchor point data for no more than 5 floors in memory at the same time.

[0064] Meanwhile, an odometry constraint edge is added between the new node and the previous keyframe node. This edge stores the relative pose transformation relationship between the two nodes (given by the inter-frame relative motion estimation output by the SLAM odometry method) and is assigned an odometry information matrix as the confidence weight of the edge.

[0065] Among them, the typical value of the odometer information matrix is ​​the value corresponding to the translation component. Rotational components correspond .

[0066] S313: During SLAM tracking, structural plane feature extraction and BIM anchor point matching are continuously performed. After a successful match, BIM anchor point constraint edges are generated, and the information matrix of the constraint edge is dynamically calculated based on the matching score and injected into the pose graph.

[0067] During the keyframe creation process, for each newly created keyframe node, the structural plane feature extraction process described in step S22 is run synchronously to extract a set of structural feature points from the point cloud of that frame.

[0068] Subsequently, the extracted structural feature points are matched with the set of BIM anchor points already loaded for the current floor. The matching algorithm uses a matching score function for scoring and filtering. When a matching pair with a score higher than the matching confidence threshold exists, the match is considered successful, and a BIM anchor point constraint edge is generated.

[0069] Connect the constraint edge between the current keyframe node and the corresponding BIM anchor node. The position of the BIM anchor node is fixed at the global coordinates determined in step S1 and remains unchanged in subsequent optimization iterations.

[0070] The information matrix of the constraint edge is dynamically calculated based on the matching score. Specifically, the matching score scaling factor is calculated. When the matching score is below 0.5, the matching score scaling factor is set to 0. When the matching score is between 0.5 and 1.0, the matching score scaling factor is linearly mapped to 0.2 to 1.0. Finally, the dynamic information matrix is ​​determined as the product of the square of the matching score scaling factor and the preset basic information matrix.

[0071] This dynamic weighting mechanism allows high-confidence anchor constraints to play a dominant role in optimization, while low-confidence constraints provide only weak guidance, and unreliable matches are automatically excluded because the scaling factor is zero.

[0072] The basic information matrix has translation components taken as follows: Rotational component This is to reflect the approximately 5 cm translation tolerance and approximately 0.5 degree angle tolerance between the BIM anchor point and the actual location on site.

[0073] S32: The hierarchical pose graph optimization algorithm driven by anchor point constraints performs progressive global optimization of the full pose graph at the local window level, floor level, and building level to obtain the global accurate pose.

[0074] The algorithm, driven by anchor point constraints, performs progressive global optimization of the entire pose graph at the local window level, floor level, and building level to obtain the global accurate pose. This includes the following sub-steps: S321: Perform high-frequency local optimization on the pose nodes within the current sliding window to instantly eliminate short-term tracking jitter; During SLAM tracking, local optimization is performed on the 25 keyframe nodes within the latest sliding window at a frequency of no less than 10Hz. Unlike subsequent levels, the core objective of this optimization layer is to quickly suppress sensor noise and local linearization errors by using BIM anchor points as absolute benchmarks while ensuring real-time performance. Therefore, incremental weighted least squares is adopted, and adaptive weights based on matching scores are introduced, so that each optimization only requires small-scale adjustments to the variables within the window.

[0075] The set of nodes to be optimized within the sliding window is The objective function is optimized as follows: , This represents the set of pose increments for all nodes to be optimized within the sliding window; the pose increment for each node is a six-dimensional vector containing three-dimensional translation corrections and three-dimensional rotation corrections. After optimization, this increment is directly superimposed on the current pose of the corresponding node. This represents the set of odometry constraint edges traversed within the sliding window, where This is the set of all odometer edges within the window, where each edge connects two temporally adjacent keyframe nodes. and ; Indicates the odometer residuals to the nodes The Jacobian matrix of the pose is 6×6; describing the nodes. The trend of the odometer edge residual when the pose changes slightly; Represents a node The pose increment vector, with a dimension of 6×1, is one of the variables to be solved in this optimization. Indicates the odometer residuals to the nodes The Jacobian matrix of the pose has a dimension of 6×6, and... Together they constitute the gradient components of the residuals for the two associated nodes; Represents a node The pose increment vector, with a dimension of 6×1, is also a variable to be solved; Represents the current residual vector of the odometry constraint edge, with a dimension of 6×1; calculated at the current pose estimate of the node, it is equal to the six-dimensional deviation between the relative transformation between the poses of the two current nodes and the odometry observation; The time-varying weight matrix representing the odometer constraint edge is a 6×6 diagonal matrix. Its diagonal elements are determined by the odometer information matrix and the reciprocal of the time interval between nodes. The shorter the time interval, the greater the corresponding weight, reflecting the characteristic of strong constraints in near frames and weak constraints in far frames. This matrix directly assigns different confidence levels to each residual component. This represents the set of BIM anchor point constraint edges that are traversed within the sliding window, where This is the set of all anchor point constraint edges within the window, with each edge connecting to a keyframe node. and a BIM anchor node ; Represents anchor point constraint edges Adaptive weighting coefficients, ,in To match the score scaling factor, The number of optimizations performed after the anchor edge is created. The decay factor is set to 0.05; this coefficient makes newly detected high-scoring anchor points contribute stronger constraints in local optimization, while the constraints of old anchor points gradually decay. This indicates the anchor point constraint residuals for nodes. The Jacobian matrix of the pose, with a dimension of 6×6, describes the effect of the node pose change on the anchor point constraint residual. This represents the current residual vector of the anchor constraint edge, with dimensions 6×1, equal to the node... The six-dimensional deviation between the current pose and the global coordinates of the corresponding BIM anchor point; The weight matrix representing the anchor point constraint edge is a 6×6 diagonal matrix. Its elements are directly assigned by the anchor point constraint information matrix. The translation and rotation components correspond to different confidence levels, reflecting the constraint strength of the BIM anchor point as an absolute spatial reference.

[0076] During the solution process, the objective function is transformed into a linear normal equation, and the pose increment is directly solved through Cholesky decomposition. The time taken for a single optimization is controlled within 10ms, and the increment is superimposed on the current pose of each node within the window to achieve real-time smoothing.

[0077] S322: When the triggering condition is met, perform unified optimization on all pose nodes within the current floor to eliminate the cumulative drift error of long-distance movement; The trigger condition for floor-level global optimization is that the cumulative number of newly created keyframe nodes in the current floor exceeds 50, or the number of newly added BIM anchor point constraint edges exceeds 5. Unlike local incremental correction, this layer aims to achieve global consistency of all nodes within a single floor. It adopts nonlinear least squares full optimization and introduces a BIM anchor point binarization screening mechanism and visual loop constraints within the floor, forming a complex optimization problem with logical judgments and multiple constraints.

[0078] Extract the current floor from the full pose graph. All pose nodes Construct the following objective function: , This represents the set of all pose nodes to be optimized within the current floor F, serving as the decision variable for this floor-level global optimization. The pose of each node is represented by a seven-dimensional vector, where the translation part is a three-dimensional vector and the rotation part is a unit quaternion. This represents the set of all odometer constraint edges within the current floor range, where This is the set of odometry edges formed between all adjacent keyframes within this floor, with each edge connecting two temporally adjacent nodes i and j; The Cauchy bar kernel function is expressed in mathematical form as follows: , where r is the squared Mahalanobis distance; this function can effectively suppress the destructive impact of severe outlier odometry edges caused by long-distance movement on the overall optimization results; The odometry edge residual function represents the six-dimensional deviation between the relative pose of node i and node j and the SLAM odometry observation, including three-dimensional translational deviation and three-dimensional rotational deviation. The information matrix representing the odometer constraint edges is a 6×6 diagonal matrix, and its translation components are taken as... Rotational component ; This represents the set of all BIM anchor point constraint edges within the current floor range, where This is the set of constraint edges formed between all keyframe nodes within this floor and their matching BIM anchor points; This represents the binarization indicator function. The matching score between node i and BIM anchor point k is given. The matching confidence threshold for floor-level optimization is set to 0.7. When the matching score is not lower than this threshold, the indicator function is set to 1, and the anchor constraint is included in the optimization; when the matching score is lower than this threshold, the indicator function is set to 0, and the anchor constraint is completely excluded. This mechanism ensures that all BIM anchors entering the floor-level global optimization are high-confidence constraints. This represents the residual function of the BIM anchor point constraint edge, which measures the current pose of node i relative to its corresponding BIM anchor point. The six-dimensional deviation between fixed global coordinates; The unified information matrix representing the anchor constraints in floor-level optimization takes a fixed value because, after being filtered by the indicator function, the anchor constraints entering this floor are considered to be high-confidence observations with equal precision. The weighted processing simplifies the complexity of the optimization problem. The weight coefficient for visual loop closure constraints is set to 0.5, which ensures that loop closure constraints serve as a beneficial supplement rather than a dominant force in the overall optimization, thus avoiding excessive interference with the overall positioning accuracy due to loop closure detection errors. This represents the set of visual loop constraint edges within the current floor range, where This is the set of loop edges detected by visual bag-of-words or point cloud feature matching within this floor, where each edge connects two spatially adjacent but temporally non-adjacent nodes; This represents the geometric consistency verification indicator function. The geometric consistency verification condition is as follows: This condition is met only when both nodes of the loop closure candidate frame pair are associated with the same or adjacent BIM anchor points. The indicator function is set to 1 and the loop closure edge is enabled; otherwise, the indicator function is set to 0 and the loop closure edge is discarded. This mechanism effectively solves the pain point of high false matching rate in traditional pure visual loop closure detection in repetitive structure scenes. The visual loop closure constraint edge residual function is used to measure the six-dimensional deviation between the relative pose of node i and node j and the loop closure detection observation. The information matrix representing the closure constraint, with the translation component taken as... Rotational component .

[0079] This optimization problem is solved using Ceres Solver, with a maximum of 30 iterations and a convergence threshold of [value missing]. After optimization, all nodes within the floor achieved internal consistency, and long-distance drift was eliminated.

[0080] S323: When a floor switching event is detected, the cross-floor constraint of the through anchor point is used to perform joint global optimization on all traversed floors and unify the cross-floor coordinate system.

[0081] When a change in the floor number of the equipment is detected and the Z-direction displacement exceeds half the floor height, a building-level global optimization is triggered. Based on the unified global coordinates of BIM anchor points, a consistency regularization term for cross-floor through-structures is introduced to explicitly penalize the horizontal translational and yaw angle deviations of the same vertical through-structure between different floors, in order to eliminate minor horizontal misalignments that may occur when each floor operates independently.

[0082] Extract all nodes of all traversed floors from the full pose graph. Construct a building-level joint optimization objective function: , This represents the set of all pose nodes to be optimized across all traversed floors, serving as the decision variable for this building-level global optimization. The pose of each node is represented by a seven-dimensional vector, where the translation part is a three-dimensional vector and the rotation part is a unit quaternion. All nodes are in the BIM global coordinate system. The set of all odometry edges formed between adjacent keyframes within the traversed floors, where each edge connects two temporally adjacent nodes i and j. The union of constraint edges formed between all keyframe nodes within a traversed floor and their matching BIM anchor points; This represents the residual function of the BIM anchor point constraint edge, which measures the current pose of node i relative to its corresponding BIM anchor point. The six-dimensional deviation between fixed global coordinates; This represents the dynamic information matrix of BIM anchor point constraint edges; the dynamic weighting mechanism is retained in building-level optimization, so that anchor point matching of different qualities contributes constraint force matching their own confidence level in global alignment.

[0083] This represents the overall weight coefficient of the cross-layer consistency regularization term, ranging from 1.5 to 3.0, with a typical value of 2.0. This coefficient ensures that the cross-layer consistency constraint has sufficient influence in the global optimization, but does not exceed the dominance of the odometry and anchor constraints. For the same vertical through-structure, a set of matching keyframe node pairs on adjacent floors is formed; for each stairwell or elevator shaft, the nearest keyframe node is found near the BIM anchor point on its adjacent floor, forming a cross-floor node pair. This represents the XY horizontal component of the 3D translation vector of node p, which is a two-dimensional vector describing the horizontal position of the node in the BIM global coordinate system; it is used to individually constrain the horizontal plane drift between floors without restricting the vertical degree of freedom. This represents the extraction of the XY horizontal components from the 3D translation vector of node q, which are related to... Two-dimensional vectors of the same dimension describe the horizontal position of corresponding nodes on adjacent floors of the same through structure in the global coordinate system; The information matrix representing the horizontal translation consistency sub-item is a 2×2 symmetric positive definite matrix. Its elements are jointly determined by the BIM design dimensions and construction tolerances of the through structure. For elevator shafts, a larger information matrix value corresponding to a smaller standard deviation (e.g., 0.02m) is used, and for stairwells, a slightly looser standard deviation (e.g., 0.05m) is used. The deviations are projected onto the principal axis direction of the structure through a rotation matrix. This information matrix reflects the regularization strategy of structural perception, that is, different through structures have different constraint stiffnesses. This represents the weighting adjustment factor of the yaw consistency sub-item relative to the horizontal translation sub-item, with a value ranging from 0.1 to 0.3, and a typical value of 0.2. Since the impact of translation error on quality is much greater than that of yaw error in building construction, this factor makes the yaw constraint play an auxiliary role in the regularization term. This represents the extracted yaw rotation component of node p about the Z-axis, describing the rotation angle of that node about the vertical direction in the global coordinate system. This operation separately constrains the yaw drift between floors, while pitch and roll angles are not within the constraints of the through-structure. This indicates the extraction of the yaw rotation component of node q about the Z-axis, and... Both are rotation angles about a vertical axis; The geodesic distance operator, representing the rotation angle, calculates the absolute value of the minimum angular difference between two rotations about the Z-axis; this operator ensures that the angular difference is always within... Within the range, avoid ambiguity caused by the periodicity of angles; The standard deviation parameter representing the yaw consistency sub-item is used, and the reciprocal of its square constitutes the information weight of this item. The value depends on the type of the through structure; for example, for stairwells, since the wall direction at the platform is strictly consistent between the upper and lower floors, it is taken as 0.5°.

[0084] The optimization uses the Ceres Solver's SPARSE_SCHUR solver, with a maximum of 50 iterations and a convergence threshold. .

[0085] After optimization, all pose nodes of the traversed floors are unified under a strict global coordinate system. The horizontal deviation of the same through structure on different floors is compressed to the allowable range of construction, resulting in a seamless global accurate pose across floors.

[0086] S4: Based on the global precise pose, overlay and render the hidden engineering component models in the BIM construction model onto the AR terminal screen; After the AR terminal device acquires the real-time accurate pose in the BIM global coordinate system, it retrieves the data of the hidden engineering components corresponding to the current construction area from the BIM model, including the routing of the pre-embedded cable pipelines in the wall, the location of water supply and drainage pipes, the cross-section of ventilation ducts, the positioning of structural embedded parts, and the size and center coordinates of reserved holes.

[0087] Based on the coordinate alignment between the current device pose and the BIM model, the 3D model of the aforementioned hidden engineering components is converted to the camera coordinate system of the AR terminal device, and after projection transformation, it is superimposed and rendered on the terminal display screen in real time in the form of semi-transparent or colored wireframes.

[0088] Construction workers can see through the wall and observe the precise spatial location of concealed engineering components through the screen. The system also displays the type name, specifications, and real-time distance from the current equipment location of each component in text labeling.

[0089] As construction workers move the equipment to different positions and angles, the rendered content updates synchronously with the equipment's pose, maintaining the spatial alignment between virtual components and the real building structure. Furthermore, the system automatically highlights the target components to be operated based on the current construction task and uses arrow guide lines to indicate the layout points or installation directions, assisting construction workers in completing precise positioning, drilling, installation of embedded parts, and quality verification operations.

[0090] S5: When a floor switching event is detected, the building-level global pose graph is optimized, and the coordinate system continuity of the AR overlay content is maintained to achieve seamless switching.

[0091] When construction workers carrying AR terminal devices enter a new floor through stairwells or elevator shafts, the system determines whether a floor switching event has occurred by measuring the change in the device's Z-axis position in the BIM global coordinate system. The determination criteria are: the current keyframe node has a different floor number attribute than the previous keyframe node, and the change in Z-axis displacement exceeds half of the design height difference between adjacent floors. Once the floor switching event is confirmed, the system automatically triggers a building-level global pose graph optimization process, performing joint global optimization on all pose nodes of traversed floors, BIM anchor point constraint edges, and cross-floor connection edges of through anchor points.

[0092] After optimization, the updated global coordinate system transformation parameters for the entire building are obtained. The spatial alignment relationship between the current device pose and the corresponding BIM anchor point in the new floor is redefined. At the same time, all pose graph nodes of the traversed floors are unified to a consistent global coordinate system.

[0093] During the optimization process, the overlay content on the AR terminal screen remains displayed without interruption. After optimization, the pose update transitions to the new pose smoothly via interpolation, avoiding screen jumps. Subsequently, the overlay rendering process for concealed engineering components continues. Construction workers can immediately obtain accurate AR assistance information on the new floor without any manual calibration or re-initialization, achieving seamless and continuous switching of construction assistance across floors.

[0094] Example 2

[0095] like Figure 2 As shown, Embodiment 2 of this application provides a construction assistance guidance system based on AR technology, including: BIM Pose Graph Construction Module 21: Used to obtain building geometry and semantic information from BIM model, establish global coordinates and calculate anchor points, and construct BIM-semantic topology pose graph; Global initial positioning module 22: used to obtain structural planar features through synchronous positioning and map building system, and combine BIM-semantic topology pose graph to perform anchor point matching and pose calculation to obtain global initial pose; Layered pose optimization module 23: used to calculate the real-time global accurate pose based on the BIM-semantic topology pose graph and the global initial pose using a layered pose graph optimization algorithm; AR overlay rendering module 24: used to overlay and render the hidden engineering component models in the BIM construction model onto the AR terminal screen based on the global precise pose; Seamless cross-floor switching module 25: When a floor switching event is detected, it triggers building-level global pose graph optimization and maintains the coordinate system continuity of the AR overlay content to perform seamless switching.

[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A construction assistance guidance method based on AR technology, characterized in that, include: By acquiring architectural geometric and semantic information from the BIM model, establishing global coordinates and calculating anchor points, a BIM-semantic topological pose diagram is constructed. The structural planar features are obtained by synchronous positioning and mapping system, and anchor point matching and pose calculation are performed by combining BIM-semantic topology pose graph to obtain global initial pose. Based on the BIM-semantic topological pose graph and the global initial pose, the real-time global accurate pose is calculated using a hierarchical pose graph optimization algorithm. Based on the global precise pose, the hidden engineering component models in the BIM construction model are superimposed and rendered onto the AR terminal screen; When a floor switching event is detected, the building-level global pose graph is optimized, and the coordinate system continuity of the AR overlay content is maintained to achieve a seamless switching.

2. The construction assistance guidance method based on AR technology according to claim 1, characterized in that, By acquiring architectural geometry and semantic information from the BIM model, establishing global coordinates and calculating anchor points, a BIM-semantic topological pose graph is constructed, including: Analyze the BIM model and extract the geometric structure information of each floor; Based on the extracted geometric information, the spatial hierarchical semantics and topological connections of the organizational structure tree are extracted; Establish a global coordinate system and calculate the global coordinates of BIM anchor points for each floor. Calculate the anchor points corresponding to the vertical passages running through different floors, and construct a BIM-semantic topology pose graph.

3. The construction assistance guidance method based on AR technology according to claim 1, characterized in that, The structural planar features are obtained through a simultaneous localization and mapping (SMR) system, and anchor point matching and pose calculation are performed using a BIM-semantic topological pose graph to obtain the global initial pose, including: The multi-sensor fusion SLAM system of the AR terminal device is activated to collect environmental data in real time and build a local point cloud map, while outputting the initial tracking pose of the device in the SLAM local coordinate system. Structural planar features are extracted from the real-time constructed local point cloud map, and the intersection lines and intersection points between the planes are calculated; The extracted structural planar features are matched online with the anchor point geometric descriptions of the current floor in the BIM-semantic topological pose map. By utilizing the successfully matched anchor point correspondence, the global initial pose of the device in the global coordinate system is obtained by solving the six-degree-of-freedom pose of the device.

4. The construction assistance guidance method based on AR technology according to claim 3, characterized in that, The multi-sensor fusion SLAM system of the AR terminal device is activated to collect environmental data in real time and construct a local point cloud map. Simultaneously, the initial tracking pose of the device in the SLAM local coordinate system is output, including: Initialize the multi-sensor data stream and complete the time synchronization and spatial extrinsic parameter calibration of the lidar, inertial measurement unit and camera; Run the laser-inertial-vision tightly coupled SLAM front-end odometry calculation method to output the device's initial tracking pose and local dense point cloud map in the SLAM local coordinate system.

5. The construction assistance guidance method based on AR technology according to claim 1, characterized in that, Based on the BIM-semantic topological pose graph and the global initial pose, a hierarchical pose graph optimization algorithm is used to calculate the real-time global accurate pose, including: The global initial pose is used as the reference node of the pose graph, and new pose nodes, odometry constraint edges and BIM anchor point constraint edges are dynamically added during the continuous SLAM tracking process to construct a real-time full pose graph that integrates BIM anchor point constraints. An anchor-constraint-driven hierarchical pose graph optimization algorithm is used to perform progressive global optimization of the full pose graph at the local window level, floor level, and building level to obtain the global accurate pose.

6. The construction assistance guidance method based on AR technology according to claim 5, characterized in that, Using the global initial pose as the reference node of the pose graph, and dynamically adding new pose nodes, odometry constraint edges, and BIM anchor point constraint edges during continuous SLAM tracking, a real-time full pose graph integrating BIM anchor point constraints is constructed, including: The global initial pose is used as the first reference node of the pose graph to initialize the full pose graph structure; As the SLAM odometry continuously outputs new poses, new pose nodes are dynamically created, and odometry constraint edges are added between adjacent nodes. During SLAM tracking, structural plane feature extraction and BIM anchor point matching are continuously performed. After a successful match, BIM anchor point constraint edges are generated, and the information matrix of the constraint edge is dynamically calculated based on the matching score and injected into the pose graph.

7. A construction assistance guidance method based on AR technology according to claim 5, characterized in that, A hierarchical pose graph optimization algorithm driven by anchor points is used to perform progressive global optimization of the entire pose graph at the local window level, floor level, and building level to obtain the global accurate pose, including: Perform high-frequency local optimization on the pose nodes within the current sliding window to instantly eliminate short-term tracking jitter; When the triggering condition is met, all pose nodes within the current floor are uniformly optimized to eliminate the cumulative drift error of long-distance movement; When a floor switching event is detected, the cross-floor constraint of the through anchor point is used to perform joint global optimization on all traversed floors and unify the cross-floor coordinate system.

8. A construction assistance and guidance system based on AR technology, characterized in that, include: The BIM pose graph construction module is used to obtain building geometry and semantic information from the BIM model, establish global coordinates and calculate anchor points to construct a BIM-semantic topology pose graph. The global initial localization module is used to obtain structural planar features through the synchronous localization and mapping system, and combine them with the BIM-semantic topological pose graph to perform anchor point matching and pose calculation to obtain the global initial pose. The hierarchical pose optimization module is used to calculate the real-time global accurate pose based on the BIM-semantic topology pose graph and the global initial pose using the hierarchical pose graph optimization algorithm. The AR overlay rendering module is used to overlay and render the hidden engineering component models in the BIM construction model onto the AR terminal screen based on the global precise pose. The seamless cross-floor switching module is used to trigger building-level global pose graph optimization when a floor switching event is detected, and to maintain the coordinate system continuity of the AR overlay content for seamless switching.

Images